Outbound email infrastructure and deliverability

A cold outbound program has two halves that fail independently. One is the list and the message. The other is the plumbing that decides whether the message arrives at all, and since February 2024 that plumbing has been governed by published rules the big mailbox providers enforce automatically. Most teams meet those rules for the first time after a domain is already cooked.

This page is the plumbing half. The strategy half lives in the B2B SaaS Outbound Marketing Playbook, and it matters more. Infrastructure only decides whether a good message gets a chance.

What the bulk sender requirements actually demand

Send more than 5,000 messages in a day to Gmail addresses and you are a bulk sender. That status brings five obligations: SPF and DKIM records on the sending domain, a DMARC policy of at least p equals none, alignment between the visible From domain and one of the authenticating domains, an RFC 8058 one-click unsubscribe header with unsubscribes processed inside two days, and a spam complaint rate held under 0.3 percent as measured in Google Postmaster Tools.

Yahoo published the same package on the same February 2024 timeline. Microsoft followed with its own version for Outlook.com, Hotmail and Live addresses from 5 May 2025, beginning with junk folder routing for non-compliant senders and tightening toward rejection after that.

The threshold is a floor, not a licence. The 0.3 percent complaint rule applies to everyone regardless of volume, and a 40-inbox cold setup that never crosses 5,000 a day to a single provider is still judged on it.

Here is the part nobody says out loud. Authentication is a weekend of DNS work and then it is done forever. Complaint rate is the hard requirement, it is decided entirely by who you send to and what you say, and no amount of infrastructure work moves it.

How to build a sending domain and inbox structure that survives

Never send cold volume from the domain your product runs on. Buy secondary domains that read as yours, redirect them at your main site, and keep your primary domain for transactional mail, product notifications and lifecycle email.

The reason is operational rather than theoretical. If a cold campaign damages your primary domain’s sender reputation, password reset emails stop arriving. That is a support incident, an on-call page and an angry customer thread, all caused by a marketing decision.

Structure follows simple arithmetic. Two to three mailboxes per secondary domain, 25 to 40 sends per mailbox per working day, five working days a week.

To sustain 2,000 cold sends a week, you need 16 warmed mailboxes across five or six domains at 25 a day. That number surprises people who budgeted for one inbox and a sequencing tool. It is also why list quality matters so much: you cannot cheaply buy your way past a bad list with more volume.

Warmup timelines and the daily caps that hold up in 2026

Warmup is four weeks of deliberately sending too little. The goal is a sending pattern that looks like a working professional rather than a machine, with replies arriving and messages being read, filed and starred.

Automated warmup pools inside tools like Instantly and Smartlead send reciprocal mail between member accounts to generate positive engagement. They help, and they carry a cost people rarely mention. Warmup traffic is counted by your own reporting and by Postmaster Tools, so your engagement metrics become a blend of real prospects and bot accounts that always open and always reply. Your dashboard looks healthy while your actual campaigns are being filtered.

If you run warmup pools, pause them for 48 hours before you take any reading you plan to act on. Otherwise you are measuring the warmup network.

List hygiene is where complaint rate is actually decided

Verification and suppression do more for deliverability than any DNS change. A verified list protects the bounce rate, and a disciplined suppression list protects the complaint rate.

Run every list through a verifier before it enters a sequence. ZeroBounce, NeverBounce and MillionVerifier all price around half a cent to one cent per address, so verifying 4,000 contacts a month costs between 20 and 40 dollars. That is the cheapest insurance in the entire program.

Then apply suppression rules that most teams only write after being burned:

• Anyone who replied negatively or unsubscribed, forever, across every sequence and every domain
• Current customers, open opportunities and anyone with a live support ticket
• Role accounts such as info, sales, support, admin and billing, which complain at far higher rates and tell you nothing about a buyer
• Contacts sourced more than 90 days ago without re-verification, since B2B contact data decays somewhere around 25 to 30 percent a year on the ranges data vendors typically quote
• Catch-all domains, or at least a separate low-volume lane for them, because verifiers cannot confirm those addresses and they usually represent 15 to 25 percent of a B2B list

One more rule that is unpopular and correct. If you cannot write the sentence explaining why this specific person would want this specific message, they do not belong on the list. That sentence is the whole difference between a 0.05 percent complaint rate and a 0.4 percent one. The outbound lead generation guide covers how to build a list that passes that test.

The diagnostic that separates deliverability from targeting

Most teams misdiagnose this, and the misdiagnosis is always in the same direction. Poor results get blamed on the inbox because infrastructure feels fixable and the message does not. Run the diagnostic before you rebuild anything.

What you observe	Most likely cause	First thing to check	Do not do this
Bounce rate above 5 percent	List quality or stale data	Re-verify the batch and check the source	Buy more domains
Bounces under 2 percent, replies near zero, seeds land in spam	Placement problem, genuine	Domain reputation in Postmaster, blacklist check, DMARC alignment	Rewrite the copy again
Bounces under 2 percent, seeds land in inbox, replies near zero	Targeting or offer	Read 50 messages aloud and ask who would answer	Add more inboxes
Replies arrive but are all negative or confused	Wrong segment entirely	Compare reply sentiment by persona and company size	Increase send volume
Every inbox on one domain drops at once	Domain level reputation event	Postmaster reputation chart, Spamhaus and SURBL lookups	Keep sending to see if it recovers
One inbox drops, others on the same domain are fine	That mailbox's sending pattern or a spam trap hit	Pause it, check its specific list segment	Blame the domain

Seed testing is the fast version. Create real mailboxes on Gmail, Outlook.com and a Microsoft 365 tenant, put them in a live campaign, and look at where the message lands. Tools like GlockApps and MailReach do this at scale. Fifteen minutes of seed testing settles the argument that otherwise runs for a month.

There is one measurement trap worth naming. Google Postmaster Tools only populates its reputation and spam rate charts once a domain sends meaningful daily volume to Gmail, and a cold setup that splits 2,000 weekly sends across six domains may never reach that volume on any single domain. The many-domains pattern that protects your reputation also removes your ability to see it. Accept the tradeoff knowingly, and lean harder on bounce rate, reply rate and seed placement as your instruments.

Why fixing infrastructure to deliver a bad message is the expensive mistake

Most teams blame deliverability for what is an irrelevant list. The tell is a team that has rebuilt its sending stack twice and rewritten its ICP zero times.

Consider the arithmetic. A team sending 2,000 messages a week at a 0.3 percent positive reply rate gets six conversations. Doubling deliverability from 60 percent inbox placement to 90 percent turns six into nine. Fixing targeting so the positive reply rate hits 2 percent turns six into forty, on the same infrastructure, with no new domains.

The second fix is harder, slower and unglamorous. It means narrowing the list, throwing away 70 percent of the accounts someone already paid to enrich, and writing a message that only makes sense to one narrow segment. Nobody gets to announce that in a standup. So teams buy more inboxes.

We spent a quarter on sending infrastructure and got our placement from decent to excellent. Reply rate moved from 0.4 to 0.5 percent. Then we cut the list from 12,000 accounts to 900 and reply rate went to 3 percent in two weeks.

That composite reflects a pattern we see repeatedly. It is also why outbound demand generation should start from segment selection and work down to plumbing, never the reverse. If you want examples of messages that earn replies rather than complaints, the cold email teardowns break down live sequences line by line.

When to stop sending entirely

Some numbers mean pause today, not optimise next sprint. Treat these as hard stops rather than warnings.

Recovery on a damaged domain takes four to six weeks of rest and a fresh warmup, and it is often not worth it. Domains cost about 12 dollars. Mailboxes cost a few dollars a month. Retiring a burned domain and standing up a clean one is usually faster and cheaper than nursing a bad reputation back, provided you have fixed the behaviour that caused it. If you have not, the new domain burns too.

There is also a strategic version of stopping. If your ACV is under roughly 8,000 dollars and your cycle is short, outbound economics rarely clear. The fully loaded cost of an SDR runs 6,000 to 9,000 dollars a month against a few hundred in infrastructure, and at low ACV the maths stops working long before deliverability does. The inbound versus outbound comparison works through where the crossover sits, and a free tool as a lead source often beats cold email at that price point.

What the infrastructure actually costs

Here is the real monthly bill for a program sending 2,000 cold messages a week, which is roughly two SDRs’ worth of volume.

Infrastructure is roughly 4 to 6 percent of program cost. The people and the data are the other 94 percent. Any hour spent tuning DKIM selectors that could have been spent narrowing the account list is a bad trade, and the ratio above explains why. This also shapes how you count results downstream, since a reply is not yet an MQL and treating it as one inflates every number above it.

What to do in the next two weeks

Start with the diagnostic, not the rebuild. Put three seed mailboxes into your live sequence today and find out where your messages land. If they land in the inbox, you have a targeting problem and no amount of DNS work will fix it.

If placement is genuinely broken, work the order that matters: verify the list, publish SPF, DKIM and DMARC on every sending domain, move cold volume off your primary domain, and rebuild warmup from week one rather than resuming at your old cap. Register every domain in Postmaster Tools while you do it so the next incident has data behind it.

Then go back to the part that decides the outcome. Narrow the segment until you can write the sentence explaining why this person wants this message, and hold the list to that standard. Everything in this guide is what makes a good message arrive. The rest of SaaS lead generation is what makes it worth arriving.